Saved in:
| Main Authors: | , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2026
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2605.16460 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866913133872807936 |
|---|---|
| author | Liu, Hui Teng, Yunlai Bai, Kunlong Qi, Pengfei Yan, Haotian Li, Liang Feng, Junlan |
| author_facet | Liu, Hui Teng, Yunlai Bai, Kunlong Qi, Pengfei Yan, Haotian Li, Liang Feng, Junlan |
| contents | Referring expression counting (REC) is an intention-driven task that requires context-aware visual reasoning. While recent vision-language models incorporate language for visual understanding, most existing REC methods rely on rulebased reinforcement learning with rewards focused primarily on final accuracy, overlooking the quality of intermediate reasoning. We propose REC-RL, a reinforcement learning framework that introduces a think-range-answer paradigm to explicitly optimize the visual reasoning process. RECRL employs Group Relative Policy Optimization and two lightweight rewards: an accuracy reward that combines range-based interval supervision with Gaussian-based precision guidance, and a format reward that enforces structured outputs. By modeling intermediate focus prediction as internal decision-making, REC-RL avoids additional annotations and better aligns with human perception. Extensive experiments demonstrate consistent improvements over strong baselines and robust generalization across benchmarks. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2605_16460 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | REC-RL: Referring expression counting via Gaussian and range-based reward optimization Liu, Hui Teng, Yunlai Bai, Kunlong Qi, Pengfei Yan, Haotian Li, Liang Feng, Junlan Computer Vision and Pattern Recognition Referring expression counting (REC) is an intention-driven task that requires context-aware visual reasoning. While recent vision-language models incorporate language for visual understanding, most existing REC methods rely on rulebased reinforcement learning with rewards focused primarily on final accuracy, overlooking the quality of intermediate reasoning. We propose REC-RL, a reinforcement learning framework that introduces a think-range-answer paradigm to explicitly optimize the visual reasoning process. RECRL employs Group Relative Policy Optimization and two lightweight rewards: an accuracy reward that combines range-based interval supervision with Gaussian-based precision guidance, and a format reward that enforces structured outputs. By modeling intermediate focus prediction as internal decision-making, REC-RL avoids additional annotations and better aligns with human perception. Extensive experiments demonstrate consistent improvements over strong baselines and robust generalization across benchmarks. |
| title | REC-RL: Referring expression counting via Gaussian and range-based reward optimization |
| topic | Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2605.16460 |